Intelligent cable quality detection system and method based on machine vision technology

Abstract

The invention relates to an intelligent cable quality detection system and method based on a machine vision technology. The intelligent cable quality detection system comprises a data acquisition unit, a self-adaptive light source unit, an image preprocessing unit, an intelligent detection unit and an early warning unit, the data acquisition unit acquires spectral data of a cable material before and after irradiation by using a multispectral imaging device, focuses on spectral reflection characteristics under different wavelengths, and acquires structural images of the material before and after irradiation by using an X-ray device; the intelligent detection unit identifies spectrum and structural characteristic changes before and after irradiation of the cable material through a material characteristic identification model to obtain material characteristic data; based on a deep convolutional neural network, automatic identification and classification of cable surface defects are carried out; and detecting the internal structure abnormity of the cable based on the 3D convolutional neural network in combination with the material characteristic data and the X-ray data. The quality control level in the cable production process is remarkably improved, and the yield and the production efficiency are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of cable quality detection, and in particular to an intelligent detection system and method for cable quality based on machine vision technology. Background Art

[0002] In the modern cable manufacturing industry, especially for the production of high-end cables using irradiation technology, it is crucial to ensure high consistency and reliability of product quality. Irradiation technology is used to improve the aging resistance and heat resistance of cable materials. Irradiation technology is usually used to change the molecular structure of polymer materials, thereby improving the aging resistance and heat resistance of cables. This process involves treating the cable with high-energy electron beams or gamma rays to form a cross-linked structure, which enables it to exhibit excellent physical and chemical properties under extreme conditions.

[0003] For irradiated high-end cables, online inspection of internal structure and surface status is crucial. The changes in material structure caused by irradiation may not be detected by naked eye observation on the surface. Therefore, machine vision and other non-destructive inspection technologies are urgently needed to provide deeper quality information. At the same time, it is necessary to overcome problems such as noise interference and surface reflection to achieve accurate identification of micro defects and potential failure points. Summary of the invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide an intelligent detection method for cable quality based on machine vision technology, especially for high-end cables with irradiation process, which can significantly improve the quality control level in the cable production process and effectively improve the yield rate and production efficiency.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] An intelligent detection system for cable quality based on machine vision technology, comprising a data acquisition unit, an adaptive light source unit, an image preprocessing unit, an intelligent detection unit, a feedback unit and an early warning unit;

[0007] The data acquisition unit includes a high-resolution industrial camera, a multispectral imaging device and an X-ray device. The multispectral imaging device is used to collect spectral data of the cable material before and after irradiation, focusing on the spectral reflectance characteristics at different wavelengths, and the X-ray device is used to obtain the structural image of the material before and after irradiation;

[0008] The adaptive light source unit automatically adjusts the angle and intensity of the light source according to the material and surface reflection characteristics of different cables;

[0009] The image preprocessing unit includes a preprocessing module and an edge detection module. The preprocessing module uses a deep denoising network to eliminate image noise and improve data quality. The edge detection module implements an improved Canny algorithm to enhance the recognition capability of low-contrast surface defects.

[0010] The intelligent detection unit is embedded with a deep convolutional neural network, a 3D convolutional neural network and a material property recognition model; the material property recognition model recognizes the changes in the spectral and structural properties of the cable material before and after irradiation to obtain material property data; the deep convolutional neural network is used to automatically identify and classify cable surface defects based on image data obtained by high-resolution industrial cameras and multi-spectral imaging equipment; the 3D convolutional neural network combines material property data and X-ray data to detect abnormalities in the internal structure of the cable;

[0011] The early warning unit automatically generates alarm information and sends it to an operator or manager.

[0012] Furthermore, the data acquisition unit includes a high-resolution industrial camera, a multispectral imaging device, and an X-ray device;

[0013] The high-resolution industrial camera acquires a high-detail image of the cable surface;

[0014] The multi-spectral imaging device acquires the reflection characteristics of the cable in multiple spectral bands, including visible light, near infrared, and ultraviolet light; uses optical filters to achieve rapid spectral switching and improve data acquisition efficiency; the multi-spectral imaging device collects spectral data at fixed intervals before and after the cable is irradiated to establish a complete spectral reflection characteristic database;

[0015] The X-ray device is used to obtain structural information inside the cable; an automatic scanning path is set to cover the full range of the cable.

[0016] Furthermore, the adaptive light source unit uses a brightness closed-loop control algorithm to adjust the brightness of the light source according to the image brightness information fed back by the high-resolution industrial camera; and automatically adjusts the incident angle of the light source according to the surface properties of the material, using a dynamic angle adjustment mechanism, driven by a stepper motor or servo motor, to accurately control the position of the light source and achieve the optimal lighting angle.

[0017] Furthermore, the deep denoising network uses DnCNN for image denoising, which effectively removes random noise in the image. The specific network architecture is as follows:

[0018] The input layer receives the original image x, and the output layer returns the denoised residual image r, so the relationship between the output and the input is in is the denoised image;

[0019] The middle layer is a deep network containing multiple convolutional layers, each of which is followed by batch normalization and ReLU activation function;

[0020] Use mean squared error as loss function to measure denoised predictions The difference with the real clean image y:

[0021]

[0022] Among them, f(x i ; θ) represents the output residual of the deep denoising network; θ is the network parameter, N is the total number of training samples; y i is the i-th target clean image; x i The i-th input sample; N is the total number of samples.

[0023] Further, the improved Canny algorithm is as follows:

[0024] According to the image I denoised by the deep denoising network, the image gradient G and direction θ are calculated using the Sobel operator;

[0025]

[0026] in, and are the gradients of image I calculated by Sobel operators in the x and y directions respectively; and the non-maximum values ​​are suppressed along the gradient direction to sharpen the edges;

[0027] Dynamically adjust high T based on local contrast high and low threshold T low , improve edge detection capabilities in low-contrast areas:

[0028] T high =α×mean(G);

[0029] T low =β×T high ;

[0030] Among them, α is the multiplication factor of the threshold, mean(G) is the average value of the gradient amplitude image, and β is a constant coefficient;

[0031] Finally, Gaussian filters of different scales are applied to detect both details and rough edge structures simultaneously, and the results of multiple scales are integrated to enhance the robustness of edge detection.

[0032] Furthermore, Gaussian filters of different scales are applied to simultaneously detect details and rough edge structures, and the results of multiple scales are combined to enhance the robustness of edge detection, as follows:

[0033] Choose a set of different scale parameters {σ1,σ2,σ3,…σ k ,...,σ K}, corresponding to different feature scales;

[0034] Perform Gaussian smoothing on the original image I at different scales to generate a set of smoothed images

[0035] Compute the gradient magnitude and direction for each smoothed image:

[0036]

[0037] Perform non-maximum suppression at each scale to maintain the local maximum in each gradient direction;

[0038] Apply variable threshold strategy to detect and connect strong and weak edges of images of various scales;

[0039] The edge strengths detected at different scales are integrated, and morphological operations are applied to eliminate small isolated points and fine pseudo edges to obtain the final edge image.

[0040] Furthermore, the material property identification model identifies the changes in the spectral and structural properties of the cable material before and after irradiation, and obtains the material property data, as follows:

[0041] Collect IR and Raman spectral data of the material before and after irradiation, remove the dryness using Savitzky-Golay filter and baseline correction, and normalize the spectral data;

[0042] Project the spectral data into the principal component space by dimensionality reduction, highlight the maximum data variability, and centralize the spectral data matrix X:

[0043] X centered =X-μ;

[0044] Where μ is the mean vector of each variable;

[0045] Calculate the covariance matrix C:

[0046]

[0047] Among them, the decomposition of eigenvalue λ and eigenvector v is performed:

[0048] Cv=λv;

[0049] Project the data into the principal component space and obtain the transformed feature matrix:

[0050] T=X centered ·P;

[0051] Where P is a transformation matrix composed of eigenvectors sorted from largest to smallest eigenvalue;

[0052] Analyze the main component loads under different irradiation conditions, identify which wave numbers correspond to loads with changes in the main components, identify changes in the characteristic intensity of specific materials, and identify new or missing characteristic peaks by comparing the main component spectra before and after irradiation;

[0053] The identified features are input into the material property identification model to generate specific material property data J.

[0054] Furthermore, the material property recognition model is obtained based on random forest training, as follows:

[0055] The model is trained using a labeled training dataset. Random forests are trained using an ensemble strategy.

[0056] Construct a training set based on input feature data and material properties, including pre- and post-irradiation states;

[0057] From the training set, randomly sample and create multiple different subsets;

[0058] Perform the following steps on each subset to build and train each decision tree:

[0059] For each node, a subset of features is randomly selected, and one feature is selected for optimal splitting. The optimal splitting criteria include information gain IG(D,j) and Gini index reduction ΔGini:

[0060]

[0061] Where D is the sample set contained in the entire training data set; x j is the jth feature; Values(x j ) is the feature x j All possible values ​​of D v By feature x j The data subset is obtained by dividing the specific value v of ; H(D) is the entropy; Gini(D) is the Gini index;

[0062] The termination criteria are set to reach the maximum tree depth, the node information gain is less than the threshold, or the minimum number of sample nodes is reached;

[0063] Use k-fold cross validation to evaluate the accuracy and stability of the model to prevent overfitting;

[0064] Use the trained decision tree model to make predictions on new spectral data to generate characteristic data and classifications describing the material state.

[0065] Furthermore, the 3D convolutional neural network combines material property data and X-ray data to detect abnormalities in the internal structure of the cable, as follows:

[0066] Collect X-ray scanning data of the material and generate three-dimensional voxel data, which is represented by a multidimensional array X∈R D ′×H×W′×C , where D′, H, W′ are depth, height and width respectively, C is the number of channels; material property data J, as an additional input feature;

[0067] The 3D convolutional neural network includes an input layer, a convolution layer, a pooling layer, a full connection layer and an output layer;

[0068] The input layer processes the extracted 3D voxel data and applies 3D convolution operations to extract spatial features;

[0069] The convolutional layer uses a 3D convolution operation to extract spatial features from the voxel data:

[0070]

[0071] Among them, W is the convolution kernel, b is the bias term, and Y (i′,j′,k′) represents the value of the output feature map at position (i′, j′, k′); W(m, n, p, q) is the weight of the convolution kernel, where m, n, p are the positions of the convolution kernel in three spatial dimensions, and q is the channel index of the input data; X (i′+m,j′+n,k′+p,q) is the value of the input voxel data at the offset position (i+m, j+n, k+p) and channel q; M, N, P are the dimensions of the convolution kernel in depth, height, and width;

[0072] The pooling layer uses 3D max pooling to reduce the data size while retaining important features:

[0073] After executing the convolution layer, a concatenation layer is added before the fully connected layer to combine the features output by the 3D convolution layer with the material property data J;

[0074] The concatenation layer works by adding the feature data as an additional input vector into the deep feature space of the neural network;

[0075] The extracted feature vector is input into the fully connected layer, high-level feature extraction is performed through a nonlinear activation function, and the output layer uses a soft maximum function for prediction.

[0076] An intelligent detection method for cable quality based on machine vision technology comprises the following steps:

[0077] According to the material and surface reflection characteristics of different cables, the angle and intensity of the light source are automatically adjusted, and the image data of the cable is obtained through a high-resolution industrial camera;

[0078] Use a deep denoising network to remove image noise and improve data quality, and implement an improved Canny algorithm to enhance the ability to identify low-contrast surface defects;

[0079] According to the processed image data, the cable surface defects are automatically identified and classified based on the deep convolutional neural network;

[0080] Use multispectral imaging equipment to collect spectral data of cable materials before and after irradiation, focusing on the spectral reflectance characteristics at different wavelengths; use X-ray equipment to obtain structural images of materials before and after irradiation;

[0081] Based on the material property identification model, the changes in the spectrum and structural properties of the cable material before and after irradiation are identified to obtain material property data;

[0082] Detection of abnormalities in the internal structure of cables using 3D convolutional neural networks combined with material property data and X-ray data;

[0083] Based on the automatic identification and classification results of cable surface defects and the detection results of cable internal structural abnormalities, alarm information is automatically generated and sent to operators or managers.

[0084] The present invention has the following beneficial effects:

[0085] The present invention;

[0086] The present invention realizes real-time analysis of changes in the spectrum and structural characteristics of cable materials before and after irradiation through a material property recognition model, obtains material property data, and uses 3D-CNN in combination with material property data to effectively detect structural anomalies inside high-end cables subjected to irradiation processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 This is a system architecture diagram of the present invention;

[0088] Figure 2 Schematic diagram of a method flow in one embodiment of the present invention. DETAILED DESCRIPTION

[0089] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0090] refer to Figure 1 , in this embodiment, an intelligent detection system for cable quality based on machine vision technology is provided, including a data acquisition unit, an adaptive light source unit, an image preprocessing unit, an intelligent detection unit, a feedback unit and an early warning unit;

[0091] The data acquisition unit includes a high-resolution industrial camera, a multispectral imaging device and an X-ray device. The multispectral imaging device is used to collect spectral data of the cable material before and after irradiation, focusing on the spectral reflectance characteristics at different wavelengths, and the X-ray device is used to obtain the structural image of the material before and after irradiation;

[0092] The adaptive light source unit automatically adjusts the angle and intensity of the light source according to the material and surface reflection characteristics of different cables;

[0093] The image preprocessing unit includes a preprocessing module and an edge detection module. The preprocessing module uses a deep denoising network to eliminate image noise and improve data quality. The edge detection module implements an improved Canny algorithm to enhance the recognition capability of low-contrast surface defects.

[0094] The intelligent detection unit is embedded with a deep convolutional neural network, a 3D convolutional neural network and a material property recognition model; the material property recognition model recognizes the changes in the spectral and structural properties of the cable material before and after irradiation to obtain material property data; the deep convolutional neural network is used to automatically identify and classify cable surface defects based on image data obtained by high-resolution industrial cameras and multi-spectral imaging equipment; the 3D convolutional neural network combines material property data and X-ray data to detect abnormalities in the internal structure of the cable;

[0095] The early warning unit automatically generates alarm information and sends it to an operator or manager.

[0096] In this implementation, the data acquisition unit includes a high-resolution industrial camera, a multispectral imaging device, and an X-ray device;

[0097] The high-resolution industrial camera acquires high-detail images of the cable surface and is equipped with adjustable focus and shutter speed settings to adapt to different shooting environments and material characteristics;

[0098] The multi-spectral imaging device acquires the reflection characteristics of the cable in multiple spectral bands, including visible light, near infrared, and ultraviolet light; key technical parameters include the ability to cover the wavelength range of 400nm to 2500nm to fully capture the spectral information of the material. Optical filters are used to achieve fast spectral switching and improve data collection efficiency; the multi-spectral imaging device collects spectral data at fixed intervals before and after cable irradiation to establish a complete spectral reflection characteristic database;

[0099] The X-ray equipment is used to obtain structural information inside the cable, including density changes and possible internal defects; key settings include voltage range (such as 50kV-100kV) and resolution (such as below 80μm); and automatic scanning path (such as spiral scanning) is set to cover the full range of the cable.

[0100] In this embodiment, the adaptive light source unit uses a brightness closed-loop control algorithm to adjust the brightness of the light source according to the image brightness information fed back by a high-resolution industrial camera; and automatically adjusts the incident angle of the light source according to the surface properties of the material (such as color, texture, and gloss), using a dynamic angle adjustment mechanism, driven by a stepper motor or servo motor, to accurately control the position of the light source and achieve the optimal lighting angle.

[0101] In this implementation, the deep denoising network uses DnCNN for image denoising, which effectively removes random noise in the image. The specific network architecture is as follows:

[0102] The input layer receives the original image x, and the output layer returns the denoised residual image r, so the relationship between the output and the input is in is the denoised image;

[0103] The middle layer is a deep network containing multiple convolutional layers, each of which is followed by batch normalization and ReLU activation function;

[0104] In this embodiment, preferably, the convolution layer configuration is:

[0105] First convolutional layer: 6464 3×33×3 convolution kernels, ReLU activation;

[0106] Several convolutional layers in the middle: 6464 3×33×3 convolution kernels, followed by batch normalization and ReLU;

[0107] The last layer: a linear convolutional layer without activation function, used to predict the residual.

[0108] Use mean squared error as loss function to measure denoised predictions The difference with the real clean image y:

[0109]

[0110] Among them, f(x i ; θ) represents the output residual of the deep denoising network; θ is the network parameter, N is the total number of training samples; y i is the i-th target clean image; x i The i-th input sample; N is the total number of samples.

[0111] The training set consists of paired clean images and images with synthetic noise; the noise model includes Gaussian noise, generated at different standard deviations. During the training process, the network parameters are updated using stochastic gradient descent, and the model performance is regularly evaluated using the validation set to prevent overfitting. The denoised image quality indicators Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) are used for evaluation.

[0112] In this implementation, the improved Canny algorithm is as follows:

[0113] According to the image I denoised by the deep denoising network, the image gradient G and direction θ are calculated using the Sobel operator;

[0114]

[0115] in, and are the gradients of image I calculated by Sobel operators in the x and y directions respectively; and the non-maximum values ​​are suppressed along the gradient direction to sharpen the edges;

[0116] Dynamically adjust high T based on local contrast high and low threshold T low , improve edge detection capabilities in low-contrast areas:

[0117] T high =α×mean(G);

[0118] T low =β×T high ;

[0119] Among them, α is the multiplication factor of the threshold, mean(G) is the average value of the gradient amplitude image, and β is a constant coefficient;

[0120] Finally, Gaussian filters of different scales are applied to detect both details and rough edge structures simultaneously, and the results of multiple scales are integrated to enhance the robustness of edge detection.

[0121] In this implementation, Gaussian filters of different scales are applied to detect both details and coarse edge structures, and the results of multiple scales are combined to enhance the robustness of edge detection, as follows:

[0122] Choose a set of different scale parameters {σ1,σ2,σ3,…σ k ,...,σ K}, corresponding to different feature scales;

[0123] Perform Gaussian smoothing on the original image I at different scales to generate a set of smoothed images

[0124] Compute the gradient magnitude and direction for each smoothed image:

[0125]

[0126] Perform non-maximum suppression at each scale to maintain the local maximum in each gradient direction;

[0127] Apply variable threshold strategy to detect and connect strong and weak edges of images of various scales;

[0128] The edge strengths detected at different scales are integrated, and morphological operations (such as opening or closing operations) are applied to eliminate small isolated points and fine pseudo edges to obtain the final edge image.

[0129] In this implementation, the material property identification model identifies the changes in the spectral and structural properties of the cable material before and after irradiation, and obtains the material property data, as follows:

[0130] Collect IR and Raman spectral data of the material before and after irradiation, remove the dryness using Savitzky-Golay filter and baseline correction, and normalize the spectral data;

[0131] Project the spectral data into the principal component space by dimensionality reduction, highlight the maximum data variability, and centralize the spectral data matrix X:

[0132] X centered =X-μ;

[0133] Where μ is the mean vector of each variable (i.e., each wave number point);

[0134] Calculate the covariance matrix C:

[0135]

[0136] Among them, the decomposition of eigenvalue λ and eigenvector v is performed:

[0137] Cv=λv;

[0138] Project the data into the principal component space and obtain the transformed feature matrix:

[0139] T=X centered ·P;

[0140] Where P is a transformation matrix composed of eigenvectors sorted from largest to smallest eigenvalue;

[0141] Analyze the principal component loads under different irradiation conditions, identify which wavenumbers correspond to loads that have changes in the principal components, identify changes in the strength of specific material features (such as chemical bonds), and identify new or missing characteristic peaks by comparing the principal component maps before and after irradiation (distribution changes are observed in the component score map);

[0142] The identified features are input into the material property identification model to generate specific material property data J.

[0143] In this implementation, the material property recognition model is obtained based on random forest training, as follows:

[0144] The model is trained using a labeled training dataset. Random forests are trained using an ensemble strategy.

[0145] Construct a training set based on input feature data and material properties, including pre- and post-irradiation states;

[0146] From the training set, randomly sample (with replacement) to create multiple different subsets;

[0147] Perform the following steps on each subset to build and train each decision tree:

[0148] For each node, a subset of features is randomly selected, and one feature is selected for optimal splitting. The optimal splitting criteria include information gain IG(D,j) and Gini index reduction ΔGini:

[0149]

[0150] Where D is the sample set contained in the entire training data set; x j is the jth feature; Values(x j ) is the feature x j All possible values ​​of D v By feature x j The data subset is obtained by dividing the specific value v of ; H(D) is the entropy; Gini(D) is the Gini index;

[0151] The termination criteria are set to reach the maximum tree depth, the node information gain is less than the threshold, or the minimum number of sample nodes is reached;

[0152] Use k-fold cross validation to evaluate the accuracy and stability of the model to prevent overfitting;

[0153] Use the trained decision tree model to make predictions on new spectral data to generate characteristic data and classifications describing the material state.

[0154] In this implementation, the 3D convolutional neural network combines material property data and X-ray data to detect abnormalities in the internal structure of the cable, as follows:

[0155] Collect X-ray scanning data of the material and generate 3D voxel data (3Dvolumetricdata), which is represented by a multidimensional array X∈R D′×H×W′×C , where D′, H, W′ are depth, height and width respectively, C is the number of channels (such as grayscale or multiple energy channels); material property data J (such as density, ductility, stress intensity, etc.) is used as an additional input feature;

[0156] The 3D convolutional neural network includes an input layer, a convolution layer, a pooling layer, a full connection layer and an output layer;

[0157] The input layer processes the extracted 3D voxel data and applies 3D convolution operations to extract spatial features;

[0158] The convolutional layer uses a 3D convolution operation to extract spatial features from the voxel data:

[0159]

[0160] Among them, W is the convolution kernel, b is the bias term, and Y (i′,j′,k′) represents the value of the output feature map at position (i′, j′, k′); W(m, n, p, q) is the weight of the convolution kernel, where m, n, p are the positions of the convolution kernel in three spatial dimensions, and q is the channel index of the input data; X (i′+m,j′+n,k′+p,q) is the value of the input voxel data at the offset position (i+m, j+n, k+p) and channel q; M, N, P are the dimensions of the convolution kernel in depth, height, and width;

[0161] The pooling layer uses 3D max pooling to reduce the data size while retaining important features:

[0162] After executing the convolution layer, a concatenation layer is added before the fully connected layer to combine the features output by the 3D convolution layer with the material property data J;

[0163] The concatenation layer works by adding the feature data as an additional input vector into the deep feature space of the neural network;

[0164] The extracted feature vector is input into the fully connected layer, high-level feature extraction is performed through a nonlinear activation function, and the output layer uses a soft maximum function for prediction.

[0165] refer to Figure 2 In this embodiment, an intelligent detection method for cable quality based on machine vision technology is also provided, comprising the following steps:

[0166] According to the material and surface reflection characteristics of different cables, the angle and intensity of the light source are automatically adjusted, and the image data of the cable is obtained through a high-resolution industrial camera;

[0167] Use a deep denoising network to remove image noise and improve data quality, and implement an improved Canny algorithm to enhance the ability to identify low-contrast surface defects;

[0168] According to the processed image data, the cable surface defects are automatically identified and classified based on the deep convolutional neural network;

[0169] Use multispectral imaging equipment to collect spectral data of cable materials before and after irradiation, focusing on the spectral reflectance characteristics at different wavelengths; use X-ray equipment to obtain structural images of materials before and after irradiation;

[0170] Based on the material property identification model, the changes in the spectrum and structural properties of the cable material before and after irradiation are identified to obtain material property data;

[0171] Detection of abnormalities in the internal structure of cables using 3D convolutional neural networks combined with material property data and X-ray data;

[0172] Based on the automatic identification and classification results of cable surface defects and the detection results of cable internal structural abnormalities, alarm information is automatically generated and sent to operators or managers.

[0173] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0174] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0175] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0176] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0177] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. An intelligent cable quality detection system based on machine vision technology, characterized in that: It includes a data acquisition unit, an adaptive light source unit, an image preprocessing unit, an intelligent detection unit, a feedback unit and an early warning unit; The data acquisition unit includes a high-resolution industrial camera, a multispectral imaging device and an X-ray device. The multispectral imaging device is used to collect spectral data of the cable material before and after irradiation, focusing on the spectral reflectance characteristics at different wavelengths, and the X-ray device is used to obtain the structural image of the material before and after irradiation; The adaptive light source unit automatically adjusts the angle and intensity of the light source according to the material and surface reflection characteristics of different cables; The image preprocessing unit includes a preprocessing module and an edge detection module. The preprocessing module uses a deep denoising network to eliminate image noise and improve data quality. The edge detection module implements an improved Canny algorithm to enhance the recognition capability of low-contrast surface defects. The intelligent detection unit is embedded with a deep convolutional neural network, a 3D convolutional neural network and a material property recognition model; the material property recognition model recognizes the changes in the spectral and structural properties of the cable material before and after irradiation to obtain material property data; the deep convolutional neural network is used to automatically identify and classify cable surface defects based on image data obtained by high-resolution industrial cameras and multi-spectral imaging equipment; the 3D convolutional neural network combines material property data and X-ray data to detect abnormalities in the internal structure of the cable; The early warning unit automatically generates alarm information and sends it to an operator or manager.

2. The intelligent detection system for cable quality based on machine vision technology according to claim 1 is characterized in that: The data acquisition unit includes a high-resolution industrial camera, a multi-spectral imaging device and an X-ray device; The high-resolution industrial camera acquires a high-detail image of the cable surface; The multi-spectral imaging device acquires the reflection characteristics of the cable in multiple spectral bands, including visible light, near infrared, and ultraviolet light; uses optical filters to achieve rapid spectral switching and improve data acquisition efficiency; the multi-spectral imaging device collects spectral data at fixed intervals before and after the cable is irradiated to establish a complete spectral reflection characteristic database; The X-ray device is used to obtain structural information inside the cable; an automatic scanning path is set to cover the full range of the cable.

3. The intelligent detection system for cable quality based on machine vision technology according to claim 1 is characterized in that: The adaptive light source unit uses a brightness closed-loop control algorithm to adjust the brightness of the light source according to the image brightness information fed back by a high-resolution industrial camera; and automatically adjusts the incident angle of the light source according to the surface properties of the material, using a dynamic angle adjustment mechanism, driven by a stepper motor or a servo motor, to accurately control the position of the light source to achieve the optimal lighting angle.

4. The intelligent detection system for cable quality based on machine vision technology according to claim 1 is characterized in that: The deep denoising network uses DnCNN for image denoising, effectively removing random noise in the image. The specific network architecture is as follows: The input layer receives the original image x, and the output layer returns the denoised residual image r, so the relationship between the output and the input is in is the denoised image; The middle layer is a deep network containing multiple convolutional layers, each of which is followed by batch normalization and ReLU activation function; Use mean squared error as loss function to measure denoised predictions The difference with the real clean image y: Among them, f(x i ; θ) represents the output residual of the deep denoising network; θ is the network parameter, N is the total number of training samples; y i is the i-th target clean image; x i The i-th input sample; N is the total number of samples.

5. The intelligent detection system for cable quality based on machine vision technology according to claim 1 is characterized in that: The improved Canny algorithm is as follows: According to the image I denoised by the deep denoising network, the image gradient G and direction θ are calculated using the Sobel operator; in, and are the gradients of image I calculated by Sobel operators in the x and y directions respectively; and the non-maximum values ​​are suppressed along the gradient direction to sharpen the edges; Dynamically adjust high T based on local contrast high and low threshold T low , improve edge detection capabilities in low-contrast areas: T high =α×mean(G); T low =β×T high ; Among them, α is the multiplication factor of the threshold, mean(G) is the average value of the gradient amplitude image, and β is a constant coefficient; Finally, Gaussian filters of different scales are applied to detect both details and rough edge structures simultaneously, and the results of multiple scales are integrated to enhance the robustness of edge detection.

6. The intelligent detection system for cable quality based on machine vision technology according to claim 5 is characterized in that: The Gaussian filters of different scales are applied to simultaneously detect details and rough edge structures, and the results of multiple scales are integrated to enhance the robustness of edge detection, as follows: Choose a set of different scale parameters {σ1,σ2,σ3,…σ k ,...,σ K }, corresponding to different feature scales; Perform Gaussian smoothing on the original image I at different scales to generate a set of smoothed images Compute the gradient magnitude and direction for each smoothed image: Perform non-maximum suppression at each scale to maintain the local maximum in each gradient direction; Apply variable threshold strategy to detect and connect strong and weak edges of images of various scales; The edge strengths detected at different scales are integrated, and morphological operations are applied to eliminate small isolated points and fine pseudo edges to obtain the final edge image.

7. The intelligent detection system for cable quality based on machine vision technology according to claim 1 is characterized in that: The material property identification model identifies the changes in the spectrum and structural properties of the cable material before and after irradiation, and obtains material property data, as follows: Collect IR and Raman spectral data of the material before and after irradiation, remove the dryness using Savitzky-Golay filter and baseline correction, and normalize the spectral data; Project the spectral data into the principal component space by dimensionality reduction, highlight the maximum data variability, and centralize the spectral data matrix X: X centered =X-μ; Where μ is the mean vector of each variable; Calculate the covariance matrix C: Among them, the decomposition of eigenvalue λ and eigenvector v is performed: Cv=λv; Project the data into the principal component space and obtain the transformed feature matrix: T=X centered ·P; Where P is a transformation matrix composed of eigenvectors sorted from largest to smallest eigenvalue; Analyze the main component loads under different irradiation conditions, identify which wave numbers correspond to loads with changes in the main components, identify changes in the characteristic intensity of specific materials, and identify new or missing characteristic peaks by comparing the main component spectra before and after irradiation; The identified features are input into the material property identification model to generate specific material property data J.

8. The intelligent detection system for cable quality based on machine vision technology according to claim 7 is characterized in that: The material property identification model is obtained based on random forest training, as follows: The model is trained using a labeled training dataset. Random forests are trained using an ensemble strategy. A training set is constructed based on input feature data and material properties, including pre- and post-irradiation states. From the training set, randomly sample and create multiple different subsets; Perform the following steps on each subset to build and train each decision tree: For each node, a subset of features is randomly selected, and one feature is selected for optimal splitting. The optimal splitting criteria include information gain IG(D,j) and Gini index reduction ΔGini: Where D is the sample set contained in the entire training data set; x j is the jth feature; Values(x j ) is the feature x j All possible values ​​of D v By feature x j The data subset is obtained by dividing the specific value v of ; H(D) is the entropy; Gini(D) is the Gini index; The termination criteria are set to reach the maximum tree depth, the node information gain is less than the threshold, or the minimum number of sample nodes is reached; Use k-fold cross validation to evaluate the accuracy and stability of the model to prevent overfitting; Use the trained decision tree model to make predictions on new spectral data to generate characteristic data and classifications describing the material state.

9. The intelligent detection system for cable quality based on machine vision technology according to claim 1 is characterized in that: The 3D convolutional neural network combines material property data and X-ray data to detect abnormalities in the internal structure of the cable, as follows: Collect X-ray scanning data of the material and generate three-dimensional voxel data, which is represented by a multidimensional array X∈R D′×H×W′×C , where D′, H, W′ are depth, height and width respectively, C is the number of channels; material property data J, as an additional input feature; The 3D convolutional neural network includes an input layer, a convolution layer, a pooling layer, a full connection layer and an output layer; The input layer processes the extracted 3D voxel data and applies 3D convolution operations to extract spatial features; The convolutional layer uses a 3D convolution operation to extract spatial features from the voxel data: Among them, W is the convolution kernel, b is the bias term, and Y (i′,j′,k′) represents the value of the output feature map at position (i′, j′, k′); W(m, n, p, q) is the weight of the convolution kernel, where m, n, p are the positions of the convolution kernel in three spatial dimensions, and q is the channel index of the input data; X (i′+m,j′+n,k′+p,q) is the value of the input voxel data at the offset position (i+m, j+n, k+p) and channel q; M, N, P are the dimensions of the convolution kernel in depth, height, and width; The pooling layer uses 3D max pooling to reduce the data size while retaining important features: After executing the convolution layer, a concatenation layer is added before the fully connected layer to combine the features output by the 3D convolution layer with the material property data J; The concatenation layer works by adding the feature data as an additional input vector into the deep feature space of the neural network; The extracted feature vector is input into the fully connected layer, high-level feature extraction is performed through a nonlinear activation function, and the output layer uses a soft maximum function for prediction.

10. An intelligent detection method for cable quality based on machine vision technology, characterized in that: The following steps are involved: According to the material and surface reflection characteristics of different cables, the angle and intensity of the light source are automatically adjusted, and the image data of the cable is obtained through a high-resolution industrial camera; Use a deep denoising network to remove image noise and improve data quality, and implement an improved Canny algorithm to enhance the ability to identify low-contrast surface defects; According to the processed image data, the cable surface defects are automatically identified and classified based on the deep convolutional neural network; Use multispectral imaging equipment to collect spectral data of cable materials before and after irradiation, focusing on the spectral reflectance characteristics at different wavelengths; use X-ray equipment to obtain structural images of materials before and after irradiation; Based on the material property identification model, the changes in the spectrum and structural properties of the cable material before and after irradiation are identified to obtain material property data; Detection of abnormalities in the internal structure of cables using 3D convolutional neural networks combined with material property data and X-ray data; Based on the automatic identification and classification results of cable surface defects and the detection results of cable internal structural abnormalities, alarm information is automatically generated and sent to operators or managers.

Images